[Bug 221073] xHCI host controller dies on resume from s2idle on AMD Strix Halo [1022:1587]

[bugzilla-daemon@kernel.org]

https://bugzilla.kernel.org/show_bug.cgi?id=221073

--- Comment #7 from Michał Pecio (michal.pecio@gmail.com) ---
Any updates? Has anyone got dynamic debug logs?

May or may not be related to bug 221103. Both issues seem to affect xHCI
controllers embedded in some generations of AMD IGPs and the controllers stop
working during or after resume from suspend, albeit with different symptoms.
The other bug has a quick and reliable repro confirmed by multiple people and
it completely hangs the machine, no recovery is known.

> @davidhubbard isolated iGPU memory allocation as a variable:
Could be due to sharing some resources with the GPU?

> This doesn't seem ideal to me if most suspends are fine.  Until we get to the
> bottom of this maybe we want a RESET_ON_FAILURE type of behavior instead to
> use in the resume routine?
Guess you're right, but I thought some people may find RESET_ON_RESUME
preferable to the status quo until the bug is solved.

The resume routine should already reset the HC if it encounters problems
(except if MMIO reads as 0xffffffff, as in bug 221103). Here resume seems to
complete successfully, but the HC doesn't function properly. Or driver state is
utterly screwed up, like suspending with a Stop Endpoint command still in
flight. Debug logs would help.

Automatic reset on Stop EP timeout or other anomalies is possible in principle
and would help in this case, but it's not implemented.

-- 
You may reply to this email to add a comment.

You are receiving this mail because:
You are watching the assignee of the bug.